The startup Higgsfield has released the 4K action comedy feature film The Cully Hill Boys, produced entirely using generative artificial intelligence in just four weeks for roughly two million dollars (42 million Czech crowns). Directed by Adilet Abish and Aitor Zoldaskali with a screenplay by Tim Planagan, the project showcases the Cinema Studio toolkit and Seedance 2.5 feature to license real celebrity likenesses without physical on-set shoots, according to CzechCrunch and The Verge.
Software Architecture and Multimodel Workflow Mechanics
Unlike traditional feature production—which typically demands budgets exceeding two million dollars and timelines stretching up to two years—Higgsfield manufactured The Cully Hill Boys over a four-week sprint. According to technical documentation and reporting from The Verge, the pipeline relied on a distributed multi-model architecture. Anthropic’s Claude handled prompt preparation, while Higgsfield’s proprietary Seedance 2.5 generated video output and synthetic speech. ByteDance’s Seedream and Google’s Nano Banana managed image and video editing.
The project served as a stress test for Higgsfield’s Cinema Studio 4.0 software. The platform is engineered to generate complete 4K environments and rendered digital avatars while preserving visual continuity across frames. According to the team’s release notes, half of the production budget was consumed purely by raw compute power. Within a single month, the system outputted over one thousand production assets and approximately one hundred distinct 4K locations.
Digital Likeness Licensing and Talent Integration
To populate the lead roles of the East London rapper narrative—adapted from Tim Planagan’s script originally submitted to the Black List platform for unproduced screenplays—the production bypassed physical casting. According to CzechCrunch and UA.News, Higgsfield licensed the digital likenesses of UFC fighter Israel Adesanya, MMA legend Quinton Jackson, content creator Matt Kiatipis, and streamer Mikyle “N3on” Rafiq.
None of the human talent spent any time on a physical film set. Instead, participants signed commercial licensing agreements, provided high-resolution photographs, and recorded approximately two minutes of audio. Software models subsequently mapped expressions, kinematics, and vocal patterns onto the digital avatars. Planagan retained the rights to traditional film adaptations of his script while granting Higgsfield generative AI rights for this specific deployment.
| Production Metric | Traditional Independent Film | The Cully Hill Boys (Higgsfield) |
|---|---|---|
| Estimated Timeline | 12 to 24 Months | 4 Weeks |
| Production Budget | two million dollars (~420M CZK) | 42 million korun |
| Primary Cost Driver | Crew salaries, travel, physical equipment | Compute power (approx. half of budget) |
| Asset Output | Iterative physical staging | 1,000+ assets across 100 4K locations |
Prompt Engineering and Visual Continuity Controls
Maintaining narrative coherence across generative video models remains a primary engineering hurdle. According to The Verge, the production team utilized explicit stylistic constraints in their prompts, referencing specific cinematic techniques by incorporating tags such as “GUY RITCHIE COVERAGE” and “Edgar Wright snap.”
To avoid chronological anachronisms in a story set in 2011, engineers hardcoded negative prompts into the algorithmic parameters, explicitly forbidding modern smartphones and newer vehicle models. For secondary and background elements, the team deployed over 140 unique digital character cards complete with fixed visual profiles and behavioral parameters. These text descriptions enforced consistent appearance and performance metrics across disparate generated cuts.
Despite algorithmic scaffolding, the production exposed persisting technical limitations inherent to current video models. Reviewers note that the film exhibits common artifacts, including illegible background text on screen interfaces and inconsistent physical interactions between characters.
Ethical Frameworks and Data Retention Policies
The project implemented strict governance to manage the privacy and intellectual property risks associated with synthetic media. According to CzechCrunch reporting, all participating celebrities reviewed and approved both the script and their digital profiles prior to production commencement.
Contractual terms mandated that all private biometric data and source recordings collected from the talent had to be permanently purged within 30 days of project completion. Higgsfield is legally barred from retaining raw biometric source files or utilizing them to train subsequent model iterations.
To foster transparency within the developer community, Higgsfield published the majority of the production prompts, character visual cards, and raw assets online via its official platform and YouTube. This open-access release enables external developers and researchers to inspect the exact prompt chains and continuity rules deployed during the four-week sprint.
